Machine learning for neural decoding
About
Despite rapid advances in machine learning tools, the majority of neural decoding approaches still use traditional methods. Modern machine learning tools, which are versatile and easy to use, have the potential to significantly improve decoding performance. This tutorial describes how to effectively apply these algorithms for typical decoding problems. We provide descriptions, best practices, and code for applying common machine learning methods, including neural networks and gradient boosting. We also provide detailed comparisons of the performance of various methods at the task of decoding spiking activity in motor cortex, somatosensory cortex, and hippocampus. Modern methods, particularly neural networks and ensembles, significantly outperform traditional approaches, such as Wiener and Kalman filters. Improving the performance of neural decoding algorithms allows neuroscientists to better understand the information contained in a neural population and can help advance engineering applications such as brain machine interfaces.
Key insights
What problem does this neural decoding tutorial address?
It shows how modern machine learning methods like neural networks and gradient boosting can replace traditional linear decoders to better predict behavior from recorded brain spiking activity.
How do modern methods compare to prior decoders?
Across monkey motor cortex, somatosensory cortex, and rat hippocampus datasets, the LSTM explained over 40% of variance left by a Wiener filter (R2 of 0.88, 0.86, 0.62 vs. 0.78, 0.75, 0.35).
What are the practical uses of these decoding gains?
Improved decoding accuracy aids brain-machine interface engineering applications and helps neuroscientists quantify information in neural populations, though all demonstrations here are offline benchmarks rather than deployed systems.
What limitations and boundaries do the authors note?
All decoding was offline, not real-time; hyperparameters may be nonoptimal; and results may not transfer to noisier modalities like fMRI or to online BMI settings with non-stationary inputs.